8 min read · Updated August 2026
How to tell if a video is a deepfake
A frame-by-frame checklist for face swaps, lip-sync fakes and fully generated clips, including what compression and re-uploads hide.
Read guideThe guides desk
Everything we've learned about telling real media from synthetic media.
Written for people who don't work in forensics. Each guide is a checklist you can run yourself with nothing but your eyes and the file, with the reasoning behind every step — including the steps that prove less than they appear to.
Start here · 7 min read · Updated August 2026
How to spot an AI-generated photo without any special tools
Image generators have fixed most of the obvious mistakes they made in 2023, but they still leave a trail. The checks below are the same ones our own analysis runs through, written so you can do them with nothing but your eyes and the file itself. Work through them in order: the early ones catch the majority of generated images in a few seconds.
Read the guideThe short version
- Start with text, hands and reflections: they catch most generated images in seconds.
- Missing metadata is a weak signal on its own; contradictory metadata is a strong one.
- Local mismatches in noise or sharpness suggest editing rather than full generation.
- Confidence comes from several independent signals agreeing, never from one score.
More from the desk
6 min read · Updated August 2026
What AI-detection tools actually measure — and where they fail
Plain-English explanation of metadata checks, compression forensics, texture statistics and model classifiers — plus the honest limits of each.
Read guide8 min read · Updated September 2026
How to tell if a voice message was cloned by AI
Voice cloning now needs seconds of sample audio. Here is what a synthetic voice gets wrong — breathing, room sound, plosives, timing — and the one check that beats all of them.
Read guide9 min read · Updated September 2026
How to check a photo from breaking news before you share it
Disaster and conflict photos spread fastest in the hours when they are least verifiable. A practical method for checking where an image really came from, using free tools.
Read guide7 min read · Updated September 2026
Spotting AI-generated photos in rentals, marketplaces and dating profiles
Generated images have moved into apartment listings, product pages and dating profiles. What to look for when money or a meeting is at stake, and the checks that settle it.
Read guide9 min read · Updated September 2026
Signal one: generation artefacts, and why models betray themselves in texture
The first of four signal types VTYAI examines. What a generation artefact actually is, where models leave them, and why they survive in some images and vanish in others.
Read guide9 min read · Updated September 2026
Signal two: physical consistency, or what light refuses to do
The second of four VTYAI signal types. How shadows, reflections, perspective and depth of field expose synthetic and composited scenes, and when physics checks mislead.
Read guide8 min read · Updated September 2026
Signal three: editing signatures, and how a real photo shows its scars
The third of four VTYAI signal types. Compression history, noise discontinuity, resampling and clone traces reveal the edited real photograph — still the most common manipulation.
Read guide8 min read · Updated September 2026
Signal four: file metadata, the evidence that is easiest to read and easiest to forge
The fourth of four VTYAI signal types. What EXIF, C2PA provenance and container data can establish about a file, why absence proves nothing, and how to read metadata without over-trusting it.
Read guideGot a file in front of you?
Run it through the checker and see which of these signals it trips. Free, no account needed.